The status of the glide in Modern Greek
Thanasis Soultatis
Abstract
This paper addresses the question of the phonemic status of the glide in Modern Standard Greek,
and it proceeds to an account for its placement and status within the Greek syllable. The
hypothesis that is proposed here is that the glide has no phonemic status in Greek. This is contrary
to the model most commonly found in the linguistic literature, which considers the glide to be
phonemic, at least in those paradigms where it does not alternate with vowels.
In this paper we advance the idea that the status of the glide is closely connected to the
stratification of the lexicon. We believe that the native Greek lexicon is divided into two strata
that exhibit distinct phonological properties. The two strata originate from the two linguistic
varieties that coexisted in Greece until 1976: an archaic, conservative variety reserved for formal
speech, and a more colloquial, low register variety, confined to the everyday speech. One
fundamental difference in the phonology of the two strata is their phonemic inventory. We use the
terms 'learned' and 'non-learned' for the two strata respectively, following a long-standing tradition
in Greek linguistic literature. We propose that in the learned stratum there is one phonemic high
front vocoid /i/ which always manifests itself as a vowel in the surface. In the non-learned
stratum, the equivalent phoneme is underspecified for the feature [±vocalic], and may therefore
surface as a [+voc] vowel or a [-voc] glide, depending on its syllabic position, phonotactic
restrictions, and the position of the lexical stress.
The model postulated here is theory-neutral and it could be easily accommodated within
frameworks like that of Optimality Theory. It offers a uniform account that is able to apply on
both verbal and nominal lexical items, belonging to both the learned and the non-learned stratum.
We do not postulate different rules, rule ordering or ad hoc effects in order to explain the
differences between the two strata. Glide syllabification is portrayed as the outcome of the
interaction of lexical rules during the derivation, which are identical for all lexical items. The only
fundamental assumption that we postulate is the existence of difference phonemic inventories for
the two strata, an assumption based on strong phonological, morphological, and historical
evidence.
Keywords: glide, syllable, coda, phoneme, stratum, level
1 Introduction
Modern Standard Greek has only one glide, the palatal j. Due to their versatile character,
glides often present exceptional phonological behaviour cross-linguistically, sometimes
patterning with vowels, sometimes with consonants. The phonemic status of the palatal glide
in Greek is disputed, with some linguists seeing it as a phoneme (Setatos, 1974, 1987;
Rytting, 2005; Nyman, 1981; Topintzi, 2011), while others have argued it is an allophone of
an underlying high vowel /i/ (Kazazis, 1968; Philippaki-Warburton, 1976). A third proposal,
that of an underspecified archi-phoneme /I/, was put forward by Malikouti-Drachman (1987)
and Malikouti-Drachman and Drachman (1990). I will present these three proposals in brief
before I move on to suggest a model which is similar, but not identical, to that of MalikoutiDrachman.
The debate on the phonemicity of the glide is not confined to Greek linguistics. The
phonemic status of the glides is a topic of debate for all languages, including English.
Authors who analyse glides as potentially phonemic include Clements and Keyser (1983),
Hayes (1989), Waksler (1990), Hume (1994), and Levi (2004, 2006). Authors who assess
glides as inherently allophonic include Steriade (1984), Kaye and Lowenstamm (1984),
UCLWPL 2013
272
Levin (1985), and Rosenthall (1994). Some authors went as far as to suggest that glides have
the same status, be it phonemic or allophonic, in all languages.
Levi (2004, 2006), categorises languages according to whether they exhibit glides or
not in their phonetic inventory, and further sub-categorises the languages that do so, into
those in which the glides are phonemic and those in which the glides are allophones of an
underlying high vowel. On the other hand, Kaye and Lowenstamm (1984) believe that
syllabification principles reply on more general cross-linguistic patterns and not any complex
and language-specific phonological rules. More precisely, they assume that syllabification is
assigned at the lexicon. Hence, the realisation of the phonemes depends on their position
within the syllable and is perfectly predictable. Syllable parsing is present in the underlying
form of morphemes, but resyllabification applies during the derivation and results in the
deletion of null constituents.
This would mean that glides are allophonic varieties of underlying vowels in nonnucleic positions. In other words, when the underlying /i/ finds itself flanked by consonants it
will surface as a vowel, while when in pre- or post-vocalic position it will surface as a
consonant, i.e., as a glide. This model seems to work for languages like English, in which we
find no phonological contrasts between high vocoids which are otherwise found in identical
melodic environments. The syllabic status of a high vocoid in English is entirely predictable
from its position in the syllable: a high vocoid in pre- or post-vocalic position has no choice
but to surface as a consonant. Thus we have words like 'yes' with a consonantal high vocoid
in onset position, but we could not have a disyllabic word like *i.es, with the initial [i]
forming a nucleus of its own.
By contrast, Greek is an example of a language in which Kaye and Lowenstamm's
model cannot be applied successfully, at least not without modifications. This is because in
Greek, unlike English, we encounter words where the high vocoid may surface as a vowel or
a consonant in an apparently arbitrary way, found in otherwise identical melodic contexts.
Indeed, there are even minimal pairs which only differ in the vocoid's realisation. Some
examples are given below:
(1)
a.
b.
c.
d.
áðia 'permission'
pión 'deeds'
viázo 'I rape'1
ípia 'mild' (f.)
áðja 'empty' (f.)
pjon 'whom' (m.)
vjázo 'I rush'
ípja 'I drunk'
Such pairs cannot be explained if we believe that the melodic position alone is
sufficient to determine the high vocoid's syllabicity. Given that these pairs differ in nothing
but the realisation of the vocoid, there are only two ways to explain the different syllabicity.
Either we will assume a phonemic glide in the words at the right, or we will accept that one
of the two groups bears some kind of marking that affects syllabification. Either way we part
from Kaye and Lowenstamm's model; admittedly more so if we accept the existence of
phonemic glides.
Although there are arguments for the inclusion of a glide in the phonemic inventory of
Greek that I will present and discuss below, I postulate the existence of an underspecified
phoneme /I/; a proposal very similar to that of Malikouti-Drachman and Drachman (1990). I
will claim that these two groups of words, presented above, constitute two phonologically
distinct strata, which together form what is Modern Standard Greek (hereafter MSG). I
1
Modern Greek lacks an infinitive, so all verbs are traditionally quoted in the 1 st person singular of the
Present tense.
UCLWPL 2013
273
suggest that this is the only solution that can explain the various realisations of the Greek
glide. All other proposals fail to explain particular nominal or verbal paradigms.
2 Distribution of the glide
2.1 Distribution in the lexicon
In order to understand the status of the glide in Greek, it is essential to understand the
stratification of the language's lexicon, which I outline below.
The MSG lexicon consists of three strata. Each has a different origin and exhibits
different phonological, morphological, and syntactic profiles. The first stratum I will call
'non-learned' (or [-learned] using a feature notation). This stratum contains items that are
found in the language 'naturally' in the sense that they have not been artificially introduced by
scholars. These items include morphemes from Ancient Greek roots and items that have been
borrowed from other languages during the history of Greek and have been successfully
incorporated into the Greek inflectional system. As well as this, the stratum contains new
words created by naive speakers – as opposed to scholars and philologists creating
neologisms ad hoc – in more recent times (the Byzantine, Ottoman, and modern eras).
As a result of their long-standing and "natural" presence in the language, these items
have undergone all the diachronic modifications determined by the phonological and
morphological rules that have been active throughout the history of Greek. These items are
also known in Greece under the term 'Demotic', which was the name of the linguistic variety
used in informal everyday situations by the native speakers before 1976, when two distinct –
albeit closely related and increasingly similar – varieties were in use in the country; the other
one being 'Katharevousa', the variety employed in formal speech.
Katharevousa was the language of written speech as well as the oral language used at
formal occasions and by most of the media. Katharevousa shared a great part of its
vocabulary with the informal Demotic, but it also included a great number of lexical items
that were artificially introduced into the language by scholars after the establishment of the
Greek state in 1830. These words were based on Ancient Greek roots and some of them were
actually invented by the scholars, such as the word ipurʝío for 'ministry', which was
composed by the ancient preposition ipó 'under' and the root erɣ, a cognate form of the
English 'work'. This word was created and inserted into the language in the 19th century in
order to replace the loanword ministérion that was in use during the first years of the newly
established Greek state.
Due to their late introduction into the language these learned lexical items skipped
some of the phonological and morphological processes that had affected non-learned words
in previous times. In fact, it may be more truthful to say that the scholars who introduced
them ignored these processes, preferring to retain a more archaic morphology and phonology.
As a result, obsolete phonological and morphological types were re-introduced into the
language and remain strong to this day.
Katharevousa and Demotic started as two quite different languages. In the 19th century
Katharevousa was barely intelligible to uneducated Greeks. However, by the middle of the
20th century, the two varieties had approached each other considerably, with Demotic
accepting many Katharevousa words and morphological types. Katharevousa as well began to
yield to an increasingly Demotic style, dropping more extreme archaic types that were too
distant from the modern language to be accepted and assimilated. By 1976 the two varieties
were so close to each other that they might have been seen as mere stylistic variations. The
official and decisive end to this situation of diglossia occurred with a decree in 1976, which
banned the use of Katharevousa altogether and declared Demotic to be the only language
UCLWPL 2013
274
used in all occasions. That Demotic of 1976 however, was already very different than the
Demotic of the beginning of the 19th century, and was actually a middle form between the
original Demotic and the original Katharevousa of 1830. As such, it accepted and retained the
greatest part of the learned lexicon, along with its special phonological and morphological
features. This lexicon comprises the second stratum of MSG.
Finally, the third stratum of the MSG lexicon includes more recent loanwords which
have not been incorporated into the native inflectional system. These words, typically from
French and English, have not been given clitic suffixes and remain monomorphemic,
violating the norm that Greek nominals and verbs are minimally bimorphemic, with one
morpheme being the inflectional suffix. Examples of this stratum are the words taksí 'taxi',
sánduits 'sandwich' from English, and kalorifér 'radiator' from French. Given the absence of
inflectional suffixes, nominals of the third stratum are not declined for case and number. 2
This stratum only includes nouns and few adjectives. Loan verbs are always given a suffix
and incorporated into the native morphology, or they are nominalised and included in
periphrastic constructions combined with the verb káno 'do', for instance káno klik 'I click',
lit. 'I do click'.
2.1.1 The glide in the stratified lexicon. The palatal glide seems to be present only in the first
stratum of Greek, that of non-learned lexical items. In these words, the palatal glide takes five
different allophonic realisations:
(a)
Word-initially or after [v, ð, b, d, z, r] it is realised as [ʝ]: [ʝatrόs] 'doctor', [vʝázome] 'I
hurry', [ðʝavázo] 'I read', [trapézʝa] 'tables', [kubʝá] 'buttons', [çérʝa] 'hands'.
After [s, t, p, f, θ] it is further devoiced to [ç]: [pços] 'who', [θçáfi] 'sulphur', [itçá]
'willow'.
After [m] it is realised as a nasal palatal stop [ɲ]: [mɲa] 'one' (fem.), [mɲázo] 'I
resemble'.
After [x, ɣ, k, g, l, n] it merges with the consonant and produces a palatal sound with
double articulation: a primary velar or coronal and a secondary palatal articulation: [x]
becomes [ç] as in [çóni] 'snow', [ɣ] becomes [ʝ] as in [arʝí] 'is late', [k] becomes [c] as in
[kacá] 'bad' (f.), [g] becomes [ɟ] as in [paráŋɟes] (plural of [paráŋga]) 'shanties', [l]
becomes [ʎ] as in [çíʎi] 'thousand' (m.), and [n] becomes [ɲ] as in [eɲá] 'nine'.
Post-vocalically it is not fricativised and it is realised as a glide [j]: [majdanós] 'parsley',
[ɣájðaros] 'donkey'.
(b)
(c)
(d)
(e)
These can be collapsed into two primary allophones. Pre-vocalically, the surface realisation is
a palatal fricative, sometimes merging with the preceding consonant, and post-vocalically a
glide. When fricativised, it assimilates in voice to the preceding consonant.
In learned lexical items the glide may only appear on the surface in fast and casual
speech, or as stylistic variation. In most cases, especially in careful speech, the high vowel [i]
retains its syllabic status in all positions regardless of stress. Thus, hiatus is much more
common in learned words than in non-learned. In iV and Vi sequences, hiatus is only
possible in [-learned] words if the high vowel is stressed. Thus, we have words like tría
‘three’ (n.) or kaíki ‘fishing boat’. When the high vowel is not stressed though, it will lose its
syllabic status and reduce to a glide. In [+learned] items though the vowel does not lose its
syllabic status even when unstressed. So, we have words like stáðio ‘stadium’, piézo ‘I push’,
2
Although some speakers will, in informal occasions or for stylistic reasons, decline some of them, usually
by adding an inflectional suffix. For instance, kalorifér would become kaloriféri with kaloriférja being its
plural.
UCLWPL 2013
275
vivlioθíki ‘library’. The vocalic character of the vocoid in these words is made obvious both
through pronunciation and the fact that they receive stress in various tokens in their
paradigms.
3 Is the glide phonemic? A review of earlier approaches
The appearance of the glide in Greek nominal and verbal paradigms can be categorised into
two groups: non-alternating glide, glides which remain as such throughout the paradigms, and
alternating j/i glide, glides that appear in some paradigm tokens only, with a vowel [i] found
elsewhere. As mentioned earlier, many authors have argued for a phonemic glide in MSG.
This is obviously easier to claim for the non-alternating glide, while in the case of alternating
j/i glides, one would have to explain the appearance of vowels in some tokens.
In the case of non-alternating glides, a phonemic status seems plausible – indeed any
other proposal would face problems and would have to answer difficult questions. One such
case is the verb paradigms that contain a pre-vocalic glide in the penultimate syllable; for
instance the verb dropjázo 'I embarrass', whose stem is /dropjaz/. The formation of the past
tense in MSG involves the use of special suffixes, as well as a stress shift to the
antepenultimate syllable. The past tense 1st person singular suffix is -a. So the past tense of
/dropjaz/ will be dropjaza.3
Applying the Antepenultimate Stress Rule (hereafter ASR), which will be discussed in
detail later, gives us the form drópjaza (as we have seen above, the glide fricativises and
devoices after [p] but for the sake of simplicity I will write it simply as <j> when its exact
pronunciation is irrelevant).
These past tense forms probably present the strongest argument for the phonemic status
of the glide. If we assume that the glide is underlyingly a vowel /i/, then the UR form of the
verb stem would be /dropiaz/, syllabified as /dro.pi.az/. Given that verb stems bear no lexical
stress (Revithiadou, 1999; Ralli, 2005; van Oostendorp, 2012), the application of the ASR
would require the stress to fall on the high vowel, which would give *dropíaza in the Past
tense. A phonemic glide would explain why ASR ignores the glide and moves on to the
previous syllable. However, as I will demonstrate later, the stress pattern of these verbs can
be explained without the need to accept an underlying glide. The same stress pattern can be
acquired through a derivational process in which syllabification and Glide Formation
precedes the ASR.
In the case of alternating j/i glides too, one could attempt to propose a phonemic glide
by suggesting the existence of different underlying forms for the same root. One case where
this suggestion can be applied is the class of neuter nouns ending in -i in the nominative
singular. This class is one of the largest in MSG and it is also quite productive. These nouns
are stressed on the penultimate or on the final syllable, and the plural is formed by the
suffixation of an -a ending. When this, or any other ending is added to them, a fricativised
glide [ʝ] appears instead of [i]. So, the noun máti 'eye' has the following paradigm:
(2)
NOMINATIVE
GENITIVE
3
SG.
SG.
Máti
matjú
PL.
PL.
mátja
matjón
This is the Imperfect tense. The same suffix and stress shift is also employed in the Simple Past tense,
which also involves the use of further suffixes. Therefore, for simplicity reasons, all mentions of the 'Past tense'
refer to the Imperfect tense.
UCLWPL 2013
276
In all tokens except for nominative singular, we encounter a glide. If we insist on a phonemic
glide for MSG we might accept that in the case of alternating j/i glides, they are allophonic
when confined to pre-vocalic positions, and retain phonemic glides in the non-alternating
paradigms only. However, the mere existence of the alternating glides, and their essentially
vocalic nature would put the idea of phonemic glides into question. It would seem to suggest
that phonemic glides developed only in environments where by accident they did not have a
chance to alternate. Where alternation did happen they still surfaced under the same phonetic
realisation but retaining their vocalic UR form.
A way to overcome this problem is to accept that the noun has two stems: /mat/ and
/matj/. Topintzi (2011) claims that the inflectional suffix decides on the base it will attach to.
Hence, the nominative singular suffix -i chooses the stem /mat/, while all other declensional
suffixes choose the stem /matj/. The reason behind Glide Formation according to Topintzi is
the preservation of the number of syllables. The stem /mat/ also appears before other affixes.
The problem with this hypothesis is that it ultimately presents exactly the same problem
mentioned in the previous paragraph. The emergence of the second stem /matj/ occurred due
to the same phenomenon, the j/i alternation, and exists because of it. The stem is proposed ad
hoc, in order to explain the alternation. Besides, the existence of a glide at the end of the
second stem does not provide any evidence that the glide is phonemic. We might as well
postulate a second stem /mati/, with the final /i/ turning into a glide when found prevocalically. This hypothesis is also not economical as it presupposes the existence of two
stems for this group of nouns.
Another problem with the two-stem hypothesis is the existence of a subgroup within
this group of -i ending neuter nouns, which end in -Cri, where C can be any obstruent. An
example is the noun alétri 'plough'. Following the two-stem model we would project a stem
/aletr/ and a stem /aletrj/. However, the latter is not manifested at all throughout the paradigm.
Exceptionally, this group of nouns shows an [i] in all tokens:
(3)
NOMINATIVE
GENITIVE
SG.
SG.
alétri
Aletriú
PL.
PL.
alétria
aletrión
This subgroup is difficult to explain using the two-stem model. We would have to accept a
subgroup with a totally different inflectional pattern than that of the rest of the group. The
subgroup would exhibit a different morphology as a result of phonological constraints (a Crj
sequence is not allowed in MSG as it creates a syllabic consonant. More about this will be
discussed later). It seems that the two-stem model can be accepted only if the final vocoid of
the second stem is underlyingly vocalic: /mati/ and not consonantal: */matj/.
Hence, I suggest that we have strong evidence that the UR form of alternating glides is
vocalic: /i/. However, I will demonstrate below that a uniform /i/ UR representation for all
MSG glides and not only the alternating ones is more explanatory and is able to account for
all realisations of the glide in a more economical way. This also avoids the problem of the ad
hoc phoneme /j/ which seems to be proposed only in order to explain the stress pattern
demonstrated earlier in verbs like dropjazo. Otherwise, the existence of this phoneme is
unmotivated and indeed counter-intuitive. It has to be a phoneme with numerous allophones
already presented earlier: [j, ʝ, ç, ɲ], or a secondary palatal articulation realised as an offglide.
All these allophones would have to be shared with another phoneme, /i/, since we have to
accept an UR /i/ in the case of alternating glides as we have seen above, which may appear as
any of these allophones within the various paradigms. Such an extensive coincidence of
allophones between two distinct phonemes is unparalleled.
Finally, another issue that emerges if we posit a phonemic glide is that the grouping of
glides into alternating and non-alternating is not always clear-cut. In some cases we have no
UCLWPL 2013
277
alternation in the inflectional paradigm, for instance in the paradigm of the noun xorjó
'village':
(4)
NOMINATIVE
GENITIVE
SG.
SG.
xorjó
Xorjú
PL.
PL.
xorjá
xorjón
In this paradigm the glide seems stable, which would lead us to accept it as phonemic. The
glide remains when other suffixes are added, like in the word xorjátis 'peasant'. However,
when this word's stem participates in compounding as the second constituent, the suffix -o is
dropped and the ending becomes -i: neoxόri 'new village'. This type, which falls beyond - or
marginally within - the inflectional paradigm shows that even when the glide is well
established in the paradigm, its representation may remain vocalic for speakers. This is
further enforced by the fact that Greek speakers generally consider this glide to be a vowel
[i], an impression that may be influenced by the spelling, which represents the glide as a
vowel. Greek spelling uses the same letter <ι> for both [i] and [j] sounds, as well as all of the
potential allophones of the glide allophones discussed above.
Philippaki-Warburton (1976) argues for phonemic /i/ in all cases and she refuses to
grant a phonemic status to the glide. The main problem with such an approach is the past
tense stress pattern we have seen above in verbs like dropjázo. If we accept a UR form
/dropiaz/ we need to explain why the past tense is drópjaza and not *dropíaza. PhilippakiWarburton believes that this is due to paradigm uniformity. She claims that the present tense
of the verb, namely dropjázo, where GF has applied and the high vocoid appears as a glide,
has led to the creation of a stem base form, to which the rest of the paradigm has to be loyal.
The present tense’s stem [dropjaz] has been upgraded to a base form, which remains
melodically unaltered in the rest of the paradigm. The past tense has to retain the segments of
the present tense and therefore it is unable to change the glide into a vowel and stress it.
Philippaki-Warburton uses the same model for nouns, for which she assumes that the base
form of the stem is that of the masculine form and it spreads to all other lexical items deriving
from the same paradigm.
Although I believe that the idea of an allophonic glide based on an underlying vowel is
correct, I fail to see how the base form is justified. One question that has to be answered is
why is it the present tense that generates the base form and not some other tense. Is there
some intrinsic quality of the present tense that gives it psychological prominence compared to
other tenses? If we accept the idea of a base form for verbs as well as for some nouns, why
this has not spread to all nouns too? Why have words like máti not generated a base form that
will be respected throughout their paradigm?
Below, I will show that a model based on lexical rules can account for all verbal and
nominal paradigms in a more uniform way, without leaving unexplained gaps and areas
where the model cannot be applied for unknown reasons, in the way Philippaki-Warburton’s
model cannot explain the absence of base form in all paradigms.
Another proposal is that by Malikouti-Drachman (1987), Malikouti-Drachman and
Drachman (1990) and Deligiorgis (1988), who argue for the existence of an underspecified
archi-phoneme /I/. In the UR form of lexical items, the archi-phoneme is unspecified for the
feature [±consonantal] and receives its value depending on the syllabic node.
4 An allophonic proposal
The model I propose here is close to that of an archi-phoneme, suggested by MalikoutiDrachman and Drachman (1990). I do not grant phonemic status to the glide but assume an
UCLWPL 2013
278
underspecified phoneme /I/, unspecified for [±vocalic] (and not for [±consonantal] as in
Malikouti-Drachman and Drachman's model). The use of the feature [±voc] provides
advantages in the description of the glides and their phonological behaviour, as demonstrated
in Nevins and Chitoran (2008).
The underspecification of the glides can explain their exceptional syllabic distribution.
Glides enjoy a largely free distribution within the syllable, which can be paralleled only to
that of the rhotic - yet the freedom of the glide surpasses that of the rhotic. Glides may follow
any consonant in a branching onset. The rhotic is also found after most consonants, but not
after stridents [s] and [z]; glides do not obey such limitation and forms tautosyllabic clusters
with stridents. The glide is devoiced and fricativised when following a voiceless obstruent
and that leads to the only cases of flat sonority voiceless branching onsets in the [-learned]
stratum. Some examples are the words θçáfi 'sulphur', and fçóŋgos 'bow'. Also after [m] the
glide strengthens to a nasal stop and that again gives the only words starting with a nasalnasal sequence in the [-learned] stratum: mɲaló 'brain', mɲázo 'I look like', mɲa 'one' (f.).
Although flat sonority onsets are not unknown in the [-learned] stratum, they are confined to
sequences of voiced fricatives, for instance in vðomáða 'week'. Elsewhere, they are only
found when the glide is the second consonant. In some more cases, the glide appears in
positions where no other consonant could be allowed. For instance, although [ft], [st], and
[str] are all allowed word-initially, the glide is the only consonant found after [ft] in words
like ftjári 'spade', and ftjáxno 'I make'. The cluster [ftr], which would be another reasonable
combination, is not attested.
This wide distribution of the glide, unlike that of the rhotic, cannot be seen as the result
of its sonority. The rhotic is found after most obstruents thanks to its sonority that is much
higher than that of the obstruents. As we have seen though, the glide surfaces as an obstruent
after obstruents, and therefore the sonority distance among them is small, or even zero. An
alternative way to explain this exceptional distribution of the glide is to attribute it to its
autosegmental profile. Glides are specified as [-cons] in the first stages of derivation when
syllable parsing occurs. They surface as fricatives or nasal stops, which means that at some
later point in derivation they switch to [+cons] through a process that we will call
obstruentisation, but this takes place after syllabification has been completed. We could
therefore examine the possibility of phonotactic constraints applying on [+cons] segments
only, while other segments being able to ignore them. This would explain why glides, albeit
obstruents in the surface, enjoy such a free distribution in onsets.
Padgett (1994) puts forward a similar proposal for Zoque, a language of Mexico. Zoque
allows no branching onsets whatsoever, but it does allow words starting with CjV sequences.
Padgett argues that this is because the glide is underlyingly vocalic. In Padgett's terms
"vocalic" is translated as [-cons].
Another way to explain this wide syllabic distribution of the post-vocalic glides would
be to place them in the nucleus. If we assume that jV sequences are nucleic, this would
explain why onset consonants' co-occurrence constraints do not apply on them. However, we
have strong reasons to believe that MSG does not allow nucleic diphthongs. Greek has both
rising and falling sonority diphthongs; in other words both jV and Vj sequences respectively.
Neither occupy the nucleus and I will present the respective arguments below. Some of them
are drawn from the work of Kaye and Lowenstamm (1984) for French and Booij (1989) for
Frisian.
Falling diphthongs (Vj):
(a)
Resyllabification of the glide: Kaye and Lowenstamm (1984) explore the syllabic
position of falling diphthongs in French. In words ending with such a diphthong, like
UCLWPL 2013
(b)
279
travail 'work', pronounced [tʁavaj], the glide resyllabifies as onset when a vowel‑initial
suffix is added: tʁa.va.je 'to work'. This could not happen if the glide was part of the
nucleus. Resyllabification of a nucleus must include the entire nucleus and not only part
of it, as Kaye and Lowenstamm demonstrated. Thus, we know that the glide in travail is
a coda. The French data is contrasted to English diphthong-final words like buy. When
a vocalic suffix is added, the glide does not resyllabify because it is part of the nucleus:
buyer is pronounced as [bɑı.ə]. In Greek, the glide does resyllabify in such examples.
Thus, in a monosyllabic word like [sόj] 'extended family', the final glide resyllabifies as
an onset when the plural suffix -a is added: so.ja. This change in syllabification is
clearly patterned by a change in pronunciation. When the glide occupies an onset
position it hardens to a palatal fricative [ʝ] and the word is pronounced [sόʝa]. In
contrast, such resyllabification and hardening do not occur in Northern Greek dialects,
something that is evidence for nucleic diphthongs.
Distributional constraints: in MSG we never encounter a diphthong before a coda. In all
cases, Vj sequences are found before an onset: maj.da.nos 'parsley', xaj.ðe.vo 'I caress'
etc. This is because MSG allows only one coda slot per syllable, which is already
occupied by the glide in these examples. Indeed, in the only case I am aware of that has
an [ai] sequence before a coda, the verb [baildízo] 'to pass out', the [ai] sequence is a
hiatus and the syllabification is ba.il.di.zo and not bajl.di.zo, which would be the case if
aj was nucleic. This can be shown if we contrast the past tense of baildízo to that of
xajðévo 'ʹto caress'ʹ, where the glide is in the coda. In the past tense the stress moves
from the penultimate to the antepenultimate in both verbs. In baildízo the stress falls on
[í]: [ba.íl.di.za], and the vocalic status of the high vowel is revealed. In xajðévo on the
other hand, the stress skips the high vocoid and moves to the low vowel: [á] and the
word is pronounced [xáj.ðe.va], evidence that the vocoid is in fact a consonant. In sum,
the presence of a coda l in baildízo blocks the high vowel from reducing to a glide,
since there is no coda slot available for it to occupy and the possibility of nucleic
diphthongs does not exist. The only option for this vowel is then to remain syllabic.
Both verbs are colloquial and not learnt so the difference between them cannot be
attributed to other factors.
Rising diphthongs (jV):
(a)
(b)
Sonority: The fact that the glide surfaces as a fricative, or even a stop in some cases
when following an obstruent, bans it from the nucleus because that would violate the
principle that determines nuclei as being more sonorous than onsets. For instance, if in
a word like [fçóŋgos] 'bow', underlyingly /fIogos/ we accept a nucleic diphthong jo,
then we would have to accept that the first part of the nucleus, namely the fricative [ç]
is of equal sonority as the onset [f].
Distributional constraints: This is probably the strongest argument against nucleic
rising diphthongs. Although we have seen previously that the glides enjoy a remarkably
free distribution, this does not mean that there are no co-occurrence constraints with the
preceding consonants. The most enlightening case is the lack of glides after rhotics. We
have already seen the case of nouns ending in -Cri, e.g., alétri 'plough' in (4) above.
They are the only neuter nouns ending in -i that do not alternate this [i] with a glide. In
addition, words starting with /rIV/ in their surface always have a vocalic [i] as the
variant of /I/, even if they are [-learned]. Essentially we see that /I/ surfaces as [i] after
the rhotic even if unstressed and followed by a vowel. This is the case only if the rhotic
and [i] are tautosyllabic. If they are heterosyllabic, as in the word xorjá 'villages', then
/I/ surfaces as a glide.
UCLWPL 2013
280
Malikouti-Drachman and Drachman (1990) explains the ban of tautosyllabic Crj sequences as
a violation of her onset structure model, which determines that a tri-consonantal onset must
be of the form Specifier-Head-Complement. If we had a stop-r-j onset, the first consonant being a stop - can only occupy the Head slot. The Specifier slot must remain vacant, since it
can only be occupied by a fricative. The only slot available then is the Complement and it is
occupied by the rhotic. This way there is no slot left for a glide and /I/ must then surface as a
vowel. This proposal does not account for word-initial /rI/ cases though. According to
Malikouti-Drachman and Drachman's model, a [rj] onset is perfectly possible: the rhotic
would occupy the Head slot and the glide would be its complement. Yet, what we see in
word-initial positions is that this type of onset is banned. Indeed, we have seen that [rj] is
always heterosyllabic in Greek, in words like xérja 'hands'.
Hall and Hamann (2010) claim that [rj], [jr] and even [ri] and [ir] sequences are
avoided cross-linguistically due to articulatory constraints. Hall and Hamann were not the
first to notice the avoidance of [rj] sequences in several languages. Other authors before
them, such as Walsh Dickey (1997), Denton (1998), and Hall (2000, 2003, 2004) had also
noticed this cross-linguistic tendency but offered various explanations.
Hall and Hamann notice that some languages ban [rj] altogether, some ban it wordinitially, and some ban it syllable-initially. The last group includes MSG, where [rj] is
prohibited if tautosyllabic but allowed if heterosyllabic. Hall and Hamann mention that [rj] is
avoided more often when tautosyllabic but they fail to provide an account for why this
happens and they adduce the lack of studies and data on the frequency of the occurrence of
the same clusters in tautosyllabic and heterosyllabic positions. They reject the possibility of
sonority playing a role in the avoidance of [rj] even though sonority could indeed offer an
explanation as to why there is a difference between heterosyllabic and tautosyllabic [rj], since
sonority is computed within syllables only. They claim that sonority actually does play a role
in heterosyllabic phonotactic restrictions too. Although sonority indeed plays a role in
heterosyllabic phonotactics - such is the case of codas that tend to be more sonorous than the
following onsets - it is also true that the acceptable coda-onset combinations are far more
varied than the combinations allowed in branching onsets, and the role of sonority in
tautosyllabic and heterosyllabic clusters cannot be equaled.
Hall and Hamann's account suffers in many aspects. It is hard to conceive why
articulatory factors would ban all instances of [rj] regardless of the phonetic nature of the
rhotic, which as they admit, may surface in several varieties: an alveolar trill, a tap, a uvular
trill, or a glide and yet it will be blocked from combining with [j]. The articulatory profiles of
all these rhotics are so different that it is impossible to think of some common articulatory
gesture that would be held responsible for the incompatibility with the glide. They also claim
that the [rj] prohibition cannot be explained through featural contour constraints because the
featural profile of glides and high vowels are identical. This does not have to be so though.
As we have seen, glides and vowels are both [-cons] but surface glides are also [-voc].
Arguments about the feature [±vocalic] and its role in the phonological behaviour of glides
the reader may find in Nevins and Chitoran (2008).
However, I agree with Hall and Hamann in that sonority cannot account for the *[rj]
constraint in MSG. Glides in MSG usually fricativise after a consonant and are pronounced
as [ʝ] or [ç]. Because fricatives are less sonorous than [r] a falling sonority RT onset emerges.
Falling sonority onsets (with the exception of [st]) are banned in many European languages,
including English, but not in Greek where we have onsets like [ft] or [xt]. A constraint
against falling sonority cannot therefore explain the ban of [rʝ] onsets. Besides, fricativisation
of the glide does not have to occur, and indeed does not always occur. The glide may be
pronounced as [j] in other contexts or even as a stop [ɲ]. In principle nothing would stop the
UCLWPL 2013
281
glide from emerging as [j] in [rj] sequences and that would solve the problem of falling
sonority: the glide is more sonorous than the rhotic and [rj] has rising sonority. Nor can we
evoke a minimal sonority requirement between the rhotic and the glide to explain their
incompatibility, because as we have seen, onsets of flat, or even falling sonority are perfectly
acceptable in learned and non-learned Greek words.
Therefore, the only way to account for this *rj constraint is to assume that MSG bans
all RC onsets, where C can be any consonant and R any liquid. 4 No segment may follow a
rhotic in the same onset and the only option for /I/ therefore is to surface as a nucleic vowel
after [r]. This constraint explains the prohibition without the need to adduce sonority and it
can explain the difference between tautosyllabic and heterosyllabic [rj] sequences. We see
that the ban of [rj] sequences in MSG can only be explained if the glide belongs to the onset
and not to the nucleus.
(c)
Monophthongisation: Nucleic diphthongs may monophthongise in certain
environments. Thus, Spanish diphthongs [ie] and [ue] monophthongise within the same
verbal paradigm when they lose the stress: [ˈxue.ɣo] becomes [xu.ˈɣar] and [ˈtie.nes]
becomes [te.ˈne.mos]. The same phenomenon occurs in French rising diphthongs,
which are nucleic in contrast with the falling ones that we discussed in the previous
paragraph. Such alternation does not occur in Greek, providing another evidence that
jV/Vj sequences are not nucleic diphthongs.
The phoneme /I/ is found only in non-learned lexical items. In [+learned] items the phoneme
is /i/, specified as [-cons, -voc] in the underlying form and it does not change its specification
during the derivation. This proposal reflects the historical events that led to the coexistence of
two separate strata within what is seen as one language. In terms of morphology and
phonology though (and to a lesser extend syntax too), the two strata function as two related
but distinct dialects. In different sociolinguistic contexts they could have been two distinct
dialects of the same language, with distinct phonological and morphological rules. Whether
or not the two strata will merge into one in the future, by sharing their phonological and
morphological properties is hard to predict, although there is evidence that [+learned] words
tend to slowly alter their morphophonology becoming more like [-learned] words. With time
some iV hiatuses in [+learned] words may be resolved and the vocoid may be realised as a
glide. Indeed this is the case with [+learned] iV sequences after dental consonants, especially
when the word is in frequent use. So, [+learned] words like ðiaforá ‘difference’ are often
realised as [ðʝaforá]. For the moment though, the stratal distinction remains robust.
Katharevousa was only abolished in 1976 and the impact of diglossia is still strong under the
superficial uniformity of MSG. I will discuss the psychological reality of the two strata in a
later section. In sum, we claim that the two strata differ in their phonemic inventories, with
the [+learned] stratum having an /i/ phoneme, while the [-learned] stratum has /I/ as its
equivalent.
As mentioned above, the phoneme /I/ recieves its specification for [±voc] depending on
its position on the melodic tier. Two factors determine its syllabicity: the segmental context
and stress. Post-vocalic /I/ receives a [-voc] specification unless other phonotactic
requirements, such as sonority distance, block it. This way two [+voc] segments in a row, i.e.,
hiatus are disfavoured in the [-learned] stratum; it is far more common in the [+learned]
stratum.
4
*RC does not apply in /lI/ sequences because the glide merges with the lateral sonorant into a palatal
lateral with double articulation: [ʎ].
UCLWPL 2013
282
One factor that can block the application of this rule is lexical stress; if the high vocoid
receives the lexical stress it will surface as [+voc]. Verbs do not have lexical stress; their
stress is assigned through rules that vary depending on the tense (Revithiadou, 1999; Ralli,
2005; van Oostendorp, 2012). Nominals though may bear lexical stress and this may fall on
any of the last three syllables of the word. An example is the word kaíki 'fishing boat', whose
stem /kaík/ bears lexical stress on the high vowel. In this case, even though the high vowel is
in post-vocalic position it is specified as [+voc].
Another case when /I/ may surface as [+voc] in pre- or post-vocalic positions is when it
would otherwise violate phonotactic restrictions. Two cases of this violation occurring exist
in MSG – at least two clear cases that cannot be explained otherwise. We have already seen
the case of pre-vocalic /I/ surfacing as [i] when it is preceded by a tautosyllabic rhotic, as in
alétri above. For instance, a [-learned] word starting with /rIV/ sequence like /rIaki/ surfaces
as [ri]: [riaki] and not *[rʝaki]. This exceptional behaviour of tautosyllabic /rI/ sequences will
be discussed in detail below.
Finally, post-vocalically, /I/ surfaces as a vowel when followed by a coda. Such words
are very rare. One example is the verb baildízo 'I pass out', which, as we have seen, is
colloquial and [-learned], never used in formal speech, and yet the high vocoid surfaces as
vocalic, albeit unstressed. This is because [l] occupies the single coda slot available in MSG.
A [-voc] glide would violate that restriction. More on the glide and codas will be discussed
later. Post-vocalic glides that are not followed by a coda acquire a [-voc] value and remain [cons, -voc] thereafter. They do not undergo obstruentisation like pre-vocalic glides do, and
they do not switch to [+cons].
A generalisation extracted from the discussion above is that only [+voc] segments may
occupy a nucleic slot in MSG. The feature [±cons] cannot be used to distinguish nucleic from
non-nucleic segments because glides in MSG only acquire a [+cons] specification when
obstruentised pre-vocalically.
4.1 Nominal vs verbal paradigms
As mentioned previously, Greek nouns may carry lexical stress. This stress may fall within
the boundaries of the morpheme or it may fall after, on the suffixes (post-accenting
morphemes, Revithiadou (1999)). Verbs on the other hand, never carry lexical stress; their
stress is determined by rules, which vary depending on the conjugation class the verb belongs
to, and of course the tense. The ASR has already been introduced; this provides an
antepenultimate stress for the Past tense of most verbs.
The specification of the feature [±voc] for /I/ during the derivation is affected by the
rules which affect the word's syllabification, namely affixation and stress assignment. In
order to explain the distribution of glides, rule ordering first needs to be explored, i.e., the
ordering of these rules: affixation, stress assignment, and determine the stage in which
syllable parsing applies. The claim here is that syllabification in MSG is cyclic and applies
throughout the derivation, every time more morphological material is added, i.e., after every
instance of affixation. However, it is not present in the underlying form, i.e., underlying
forms are not syllabified prior to Level 1 affixation. Stress assignment applies after Level 2
affixation.
More precisely, there are three types of affixes in Greek: Level 1 affixes, which include
the derivational suffixes; Level 2 affixes, which include inflectional suffixes, and Level 3
affixes, which include the augment, and prefixes such as ek- and ev-. Level 1 affixes affect
stress, while those of levels 2 and 3 do not.
In the tableaux below I present a suggestion for the syllabification of words – both
verbs and nouns – that contain a high vocoid /I/ or /i/.
UCLWPL 2013
283
Tableau A: [-learned] verbs
The first and the third verbs are in the Past tense and therefore receive antepenultimate stress,
while the second is in the Present tense and is monosyllabic. There are no derivational affixes
added onto these verbs, so no affixation occurs in Level 1. In Level 2 the verbs acquire their
inflectional suffixes. Syllable parsing follows immediately after.
épjana 'I cought'
UR
[[pIan]a]
L1 affixation
L2 affixation
[pIana]
Syllabification
[pja.na]
Stress assignment L3 affixation
[é[pja.na]]
Brackets erasure [épja.na]
Syllabification
[é.pja.na]
Obstruentisation [é.pça.na]
pjo 'I drink'
[[pI]o]
[pIo]
[pjo]
[pjό]
[pjó]
[pço]
drópjaza 'I embarrassed'
[[dropIaz]a]
[dropIaza]
[dro.pja.za]
[dró.pja.za]
[dró.pça.za]
As the Past tense requires antepenultimate stress and [pja.na] is bisyllabic, stress fails to be
assigned in [pja.na] after Level 2 affixation. Therefore, the augment e- is added in Level 3
and receives the stress. No such stress rule applies in dropjaza as it does have an
antepenultimate syllable, or in pjo because it is in the Present tense.
Tableau B: [+learned] verbs
The verb shown is in the Past tense and therefore receives antepenultimate stress. However,
as it is trisyllabic, there is no need for the augment to be prefixed and no material is added in
L3. The high vocoid is pre-specified as [+voc] and it must therefore obtain a syllabic status.
píeza 'I pushed'
UR
[[piez]a]
L1 affixation
[pieza]
Syllabification
[pi.e.za]
L2 affixation
Syllabification
Stress assignment [pí.e.za]
Tableau C: nominals
Three nominals are shown: two nouns and one adjective. The first noun is [-learned], the
second is [+learned] and therefore they contain different phonemes, /I/ and /i/ respectively.
The adjective is given as an example of a word containing an L1 affix. This affix is postaccenting, i.e., it requires the stress to fall on its right. No morphological material is added in
L3 in any of the three words.
UR
L1 affixation
Syllabification
L2 affixation
Syllabication
xorjó 'village' (-learned)
[xorI]ó]
[xorIó]
[xor.jó]
stáðio 'stadium' (+learned)
[[stáði]ó]
[stáðió]
[stá.ði.ó]
majikós 'magic'
[[[máɣ]ik´]ós]
[[máɣik´]ós]
[[má.ɣik´]ós]
[má.ɣik´.ós]
[má.ɣi.k´ós]
UCLWPL 2013
Stress assignment
L3 affixation
Syllabification
Obstruentisation
284
[xor.ʝó]
[stá.ði.o]
-
[ma.ɣi.kós]
-
The stem [xorI] bears no lexical stress, so the only morpheme carrying stress is the
inflectional suffix. Therefore, this stress surfaces as it faces no competition. The stem [stáði]
carries lexical stress on its initial syllable however. The stem's stress clashes with the suffix's
stress, but the stem wins as it is the head of the morphological structure (Revithiadou, 1999).
In the case of the adjective, the stem carries a lexical stress and so does the derivational suffix
[ik´], though this latter stress is floating: [ik´] is a post-accenting suffix that requires the stress
to fall on its right (Revithiadou, 1999; Ralli, 2005; van Oostendorp, 2012). So, once the
inflectional suffix is added to its right, it receives the floating stress. Out of the three stresses
(the stem's, the derivational suffix's, and the inflectional suffix's), the derivational suffix's
stress wins, as this is the head of the morphological structure.
5 The post-vocalic glide
Apart from its exceptionally wide distribution within onsets, another important observation
with regards to the glide's positioning in the syllable is that it may occupy the unique coda
slot. Although they are not numerous, there are words that have CVj syllables – clearly all
these words belong to the [-learned] stratum. Such examples include the words majdanós
'parsley', ɣájðaros 'donkey', ajðóni 'nightingale', tsáj 'tea', xajðévo 'I caress', and korojðévo 'I
make fun of'. We know that these glides occupy the coda slot as they are never found before
another coda. We saw the example of the verb baildízo earlier, which is clearly [-learned] and
yet the /I/ phoneme fails to receive the [-voc] specification because it is followed by the coda
[l].
This is another piece of evidence that Greek is not an open syllable language as has
been claimed by some authors (Setatos, 1974; Malikouti-Drachman, 1984; Kappa, 1996). The
glide along with the liquids can occupy coda positions word-medially. As most of these
words are actually recent acquisitions in the language (Babiniotis, 2010), they constitute a
case of coda emergence in recent periods of the language's history, which is contrary to the
belief that through the course of time Greek has gradually dropped codas and moved towards
an open syllable pattern.
6 The strata of Greek, the feature [±learned], and its psychological reality
A word must be said about the reality and the status of lexicon stratification in Modern Greek
today. In most of the literature (Kappa, 1996; Simeonidi & Fliatouras, 2004; Rytting, 2005;
Topintzi, 2011) the two native strata are suggested to be marked with the feature [±learned].
Lexical items that have been introduced to MSG through Katharevousa are thought to be
specified as [+learned], while words that have been in the everyday lexicon “naturally” are
marked as [-learned]. The use of the term ‘learned’ is disputable as [+learned] words are not
acquired by speakers in any different way than the [-learned] ones. The strata are both
acquired by infants through exposure to them in their ambient environment and school,
although it is probably true that more [+learned] words are learned through education than [learned] ones, which are more commonly learned from the interaction with family and other
people in everyday context. Philippaki-Warburton (1976) uses the feature [±kath] from the
UCLWPL 2013
285
previous state-language, Katharevousa. Philippaki-Warburton’s paper was published in the
year Katharevousa was abolished and it was likely written when Katharevousa was still the
state language. This feature is even more distant from today’s linguistic reality than
[±learned]. Katharevousa has now been out of use for thirty-seven years and speakers
younger than forty years old have had practically very little - if any - experience of diglossia.
It is hard to imagine how these thousands of words that are perfectly incorporated into the
everyday lexicon are seen today as belonging to a linguistic variety that is now effectively
‘dead’.
Rytting (2005) has carried out a number of experiments asking native speakers to assess
lexical items by categorising them by whether they think of them as suitable for formal
conversations or suitable for informal conversations only. Participants were given lists of
words from various semantic affiliations, names of objects as well as more abstract meanings,
both rare and common. Rytting concluded that the words of the two strata differ today in their
formality. Words of the [+learned] stratum are suitable for formal conversations, talks etc.
while [-learned] words are confined to informal occasions, such as discussions with close
friends and family.
It is true that words of the [+learned] stratum are suitable for formal talk. On the other
hand, it would be wrong to claim that [-learned] words are not. In numerous cases [-learned]
words have no [+learned] substitutes for use in formal conversations. Common nouns such as
‘child’, ‘hand’ and verbs such as ‘to sit’ are such examples and yet they are used without
objection in formal speech. It is also true that many [-learned] words are often considered to
be too vulgar to be used in formal register, for instance in the media, and they are substituted
with more ‘formal’, literary words, even if these words are obsolete and very rarely – if ever
– used in everyday language. Examples include names of animals, such as ‘pig’, ‘hen’ etc.
whose colloquial names are normally avoided in formal occasions and replaced with the
Katharevousa words, even though they are not used in everyday conversations and they are
therefore not part of the MSG lexicon.
These words are indeed "learned", in the literal sense and they must be contrasted with
words that originated in Katharevousa but have been successfully accepted by speakers, are
commonly used and form part of the MSG vocabulary – indeed often having no [-learned]
alternative. Such words can be so common, such as high frequency words like ‘room’ or
‘book’, that it is hard to imagine them being any different than [-learned] words in their
psychological reality.
In summary, the current situation for MSG is quite fluid. Although there is a great
number of word pairs with the same meaning, differing in their value for [±learned] and used
in different occasions depending on the degree of formality, there are many [-learned] words
that are commonly used in formal situations and many [+learned] words that have
successfully made their way to even the most informal conversations. As a result of this, it is
difficult to clearly see how all words originating from Katharevousa form a distinctive set in
the mind of speakers. Renaming the feature to [±formal] might bring us closer to the real
status of many words, but it will still fail to provide some clear cut stratum division. Rytting’s
experiments are also unable to provide evidence for such a clear distinction as they rely on a
relatively small number of words. As the degree formality generates a formality continuum
rather than a clear-cut two-way division, it is particularly difficult to assess the formality
status of the entire vocabulary with such experiments.
It seems that the distinction between [+learned] and [-learned] lexical items today is
one of different grammars rather than of different psychological realities. The two strata still
exist today, although not clearly separated and with many words falling somewhere in the
middle. However, they only reflect grammar differences, i.e., different phonology,
morphology and to a lesser degree syntax. It should also be noted that an increasing number
UCLWPL 2013
286
of [+learned] items join the [-learned] stratum, in a slow but steady process, which may need
centuries to be completed. So, many words with an iV hiatus undergo GF – at least in
informal speech – and generate jV sequences instead. Verbal morphology is particularly
dynamic in MSG and [+learned] structures gradually give way to [-learned] ones. So, verbs
whose Present tense ended in –o, now have an acceptable –ao alternative, for instance oðiɣo
'I drive' is more and more realised as oðiɣao, while the colloquial verbal ending –epso is
often found replacing the [+learned] –efso. The situation can be compared to Latinized words
in English, or the stratification of the Japanese lexicon (Itō & Mester, 1999) which obey
different phonological principles but they cannot be claimed to bear a psychological mark.
Therefore, the use of the term ‘learned’ in this paper is purely a convention.
7 Conclusion
This paper argues for a stratification of the Greek lexicon arising as a result of historical and
artificial factors that affected the language in previous times, but which failed to create a
long-lasting distinction in the psychological representation of the strata. The Modern Greek
lexicon comprises of two native and one ‘non-native’ stratum, which differ in terms of the
phonological, morphological, and syntactic principles they obey.
Amongst the differences in the Phonology of the two native strata, a difference should
be added in their phonemic inventory. The [+learned] stratum contains an /i/ phoneme, prespecified as [-cons, +voc], while the [-learned] stratum contains an /I/ phoneme instead,
which is specified as [-cons] but unspecified for the feature [±vocalic]; it is specified during
the derivation according to the syllabic node in which the phoneme finds itself. Pre- or postvocalic /I/ surfaces as a [-cons, -voc] glide and is usually further strengthened to a fricative.
However, when it is found between consonants or when it receives lexical stress, /I/ surfaces
as a [-cons, +voc] high vowel [i] and constitutes a syllabic nucleus. The glide in [-learned]
items enjoys a remarkable freedom in distribution and can form tri-consonantal onsets as well
as occupy the coda position.
The idea of a different phoneme in the two strata has not been put forward in the
literature before. Yet, there is nothing to stop us from assuming that the “forced” introduction
of so many lexical items led to the introduction of a new phoneme too, the phoneme /i/. With
regards to its phonetic realisation [i], this sound was already part of the Greek phonetic
inventory, as an allophone of the phoneme /I/. The absence of an allophonic alternation
depending on the syllabic position, as happens in the [-learned] stratum, resulted in the
phonemicisation of the [i] allophone to a new phoneme in these [+learned] words.
Demonstrated here is an account based on an underspecified phoneme /I/ being
successful in accounting for all the instantiations of the glide, as well as the vowel/glide
alternations within paradigms. These are hard to explain using models based on a phonemic
glide /j/, or a vocalic phoneme /i/. The underspecified phoneme proposal offers a unified,
straight-forward, and economical account. It does not need to include problematic
suggestions found in earlier literature, such as paradigm uniformity effects (PhilippakiWarburton, 1976), ad hoc allomorphy (Topintzi, 2011), abstract articulatory constraints (Hall
& Hamann, 2010), and it can also account for exceptional cases like the –Cri sequences,
which have been often ignored in the literature.
As the idea of a phonemic glide is rejected for MSG, it is necessary to explore the
processes and contrasts within the autosegmental profile of the high vocoids, which can
account for their different phonotactic behaviour. We assumed therefore, that despite the lack
of a phonemic glide, the difference between a surface vowel and a glide is essentially autosegmental. Glides are specified as [-vocalic], while vowels as [+vocalic]. The [-voc]
UCLWPL 2013
287
specification of the glide however does not originate in the lexicon, as it does in phonemic
models, but it is acquired during the derivation. Both glides and front high vowels may
emanate from the same unspecified phoneme /I/. Clearly, we assume that the feature [±voc],
which has been considered to be redundant in much of the recent linguistic theory, is
necessary in order to explain the different behaviour between glides and vowels.
References
Anastasiadi-Simeonidi, A., & Fliatouras, A. (2004). Η Διάκριση [λόγιο] και [λαϊκό] στην Ελληνική Γλώσσα,
Ορισμός και Ταξινόμηση [The [learned] and [colloquial] distinction in Greek, Definition and
Classification]. Paper presented at the 6th International Conference on Greek Language.
Babiniotis, G. (2010). Ετυμολογικό Λεξικό της Νέας Ελληνικής Γλώσσας [Etymological Dictionary of Modern
Greek]. Athens: Centre for Lexicology.
Booij, G. (1989). On the representation of diphthongs in Frisian. Journal of Linguistics, 25, 319-332.
Clements, G. N., & Keyser, S. J. (1983). CV Phonology: A generative theory of the syllable. Cambridge, MA:
MIT Press.
Deligiorgis, I. (1988). Glides and syllables (Unpublished PhD thesis). University of Iowa, United States.
Deligiorgis, I. (1993). The status of [consonantal]. Cahiers de linguistique théorique et appliquée, 30, 3-15.
Denton, J. M. (1998). Phonetic perspectives on West Germanic Consonant Gemination. American Journal of
Germanic Linguistics and Literatures, 10, 201-235.
Hall, T. A. (2000). Typological generalizations concerning secondary palatalization. Lingua, 110, 1-25.
Hall, T. A. (2003). Phonetics in phonology: The markedness of rhotic + glide sequences in English. Folia
Linguistica, 37, 249-267.
Hall, T. A. (2004). On the non-gemination in West Germanic twenty-one years later. Folia Linguistica Historica,
25, 211-234.
Hall, T. A., & Hamann, S. (2010). On the cross-linguistic avoidance of rhotic plus high front vocoid sequences,
Lingua, 120, 1821-1844.
Hayes, B. (1989). Compensatory lengthening in moraic phonology. Linguistic Inquiry, 20(2), 253-306.
Hume, E. (1994). Representing the duality of glides. Paper presented at Langues et Grammaire, Université de
Paris 8, France.
Itō, J., & Mester, A. (1999). The phonological lexicon. In N. Tsujimura (Ed.), The Handbook of Japanese
Linguistics. Blackwell.
Kappa, I. (1996). Συλλαβοποίηση στην Κοινή Νέα Ελληνική [Syllabification in Modern Standard Greek].
Studies in Greek Linguistics, 16, 83-91.
Kaye, J. D., & Lowenstamm, J. (1984). De la syllabicité [On Syllabicity]. In F. Dell, D. Hirst & J. R. Vergnaud
(Eds.), Forme sonore du langage (pp. 123-160). Paris: Hermann.
Kazazis, K. (1968). Sunday Greek. Journal of Modern Greek Studies, 10, 57-69.
Levi, S. (2004). Underlying glides vs. vowels: typology and case studies. Paper presented at the 3rd North
American Phonology Conference (NAPhC3), United States..
Retrieved from https://files.nyu.edu/svl2/public/papers/NAPHC3_typology.pdf
Levi, S. V. (2006). Two types of glides: phonological or phonetic differences? Paper presented at LSA
Workshop: Phonetics and Phonology of Glides Albuquerque, New Mexico, United States.
Retrieved from https://files.nyu.edu/svl2/public/papers/LSA06_glide_workshop_HO.pdf
Levin, J. (1985). A metrical theory of syllabicity (Unpublished PhD thesis). MIT, Cambridge, MA, United
States.
Malikouti-Drachman, A. (1984). Syllables in Modern Greek. In W. U. Dressler, H. C. Luschützky, O. E.
Pfeiffer & J. R. Rennison (Eds.), Phonologica (pp. 181-186). Cambridge: Cambridge University Press.
Malikouti-Drachman, A. (1987). Η αντιπροσώπευση των διπλών συμφώνων στα Νέα Ελληνικά, Μελέτες για την
Ελληνική Γλώσσα, Πρακτικά της 8ης ετήσιας συνάντησης του Τομέα Γλωσσολογίας του ΑΠΘ [The
representation of geminates in Modern Greek]. Studies on Greek Language, Proceedings of the 8th Annual
Meeting of the Linguistics Department of the Aristotle University of Thessaloniki, 275-293.
Malikouti-Drachman, A., & Drachman, G. (1990). Φωνολογική Κυβέρνηση και Προβολή: Αφομοιώσεις,
Ανομοιώσεις [Phonological government and projection: Assimilations, dissimilations]. Working Papers in
Greek Grammar, University of Salzburg, 1-20.
Malikouti-Drachman, A. (2001). Greek phonology: a contemporary perspective. Journal of Greek Linguistics 2,
187-243.
UCLWPL 2013
288
Nevins, A., & Chitoran I. (2008). Phonological representations and the variable pattering of glides. Lingua,
118(2), 1979-1997.
Nyman, M. (1981). Paradigms and trans-derivational constraints: Stress and yod in Modern Greek. Journal of
Linguistics, 17(2), 231-246.
Padgett, J. (1994). Stricture and nasal assimilation. Natural Language and Linguistic Theory, 12(3), 465-513.
Philippaki-Warburton, I. (1976). On the boundaries of morphology and phonology: a case study from Greek.
Journal of Greek Linguistics, 12, 259-278.
Ralli, A. (2005). Μορφολογία [Morphology]. Athens: Pattakis.
Revithiadou, A. (1999). Headmost accent wins, head dominance and ideal prosodic form in lexical accent
systems (Unpublished PhD thesis). University of Leiden, Netherlands.
Rosenthall, S. (1994). Vowel/glide alternation in a theory of constraint interaction (Unpublished PhD thesis).
University of Massachusetts, Amherst, United States.
Rytting, A. C. (2005). An iota of difference: Attitudes to yod in lexical and social contexts. Journal of Greek
Linguistics, 6(1), 151-185.
Setatos, M. (1974). Φωνολογία της Κοινής Νεοελληνικής [Phonology of Modern Standard Greek]. Athens:
Papazisis.
Setatos, M. (1987). Από τα αρχαία στα νέα ελληνικά: Φωνολογικές αλλαγές, Μελέτες για την Ελληνική Γλώσσα,
Πρακτικά της 8ης ετήσιας συνάντησης του Τομέα Γλωσσολογίας ΑΠΘ [From ancient to modern Greek:
Phonological changes]. Studies on Greek Language, Proceedings of the 8th Annual Meeting of the
Department of Linguistics of the Aristotle University of Thessaloniki, 187-193.
Topintzi, N. (2011). Paradigm structure and Greek glides. Paper presented at University of Leipzig, Germany.
Retrieved from http://www.uni leipzig.de/~topintzi/papers/leipzig.pdf
van Oostendord, M. (2012). Stress as a proclitic in Modern Greek. Lingua, 122(11), 1165-1181.
Waksler, R. (1990). A formal account of glide/vowel alternation in prosodic theory (Unpublished PhD thesis).
Harvard University, United States.
Walsh Dickey, L. (1997). The phonology of liquids (Unpublished PhD thesis). University of Massachusetts,
Amherst, United States.